Middleware 2003 ACM/IFIP/USENIX International Middleware Conference Rio Othon Palace Hotel Rio de Janeiro, Brazil 16-20 June 2003

[ Home ] [ Program ] [ Registration ] [ Organization ] [ Sponsors ] [ CFP ] [ Important Dates ] [ Submissions ] [ Travel ]

Tutorials
Keynote Speech
Posters
Workshops
Work in Progress
Student Travel Grants
Student Volunteer Program

Website Mirror at the University of São Paulo

Full Paper Abstracts

Content Distribution for Publish/Subscribe Services

Mao Chen, Andrea LaPaugh and Jaswinder Pal Singh (Princeton U.)

Publish/subscribe systems have focused primarily on matching events to subscriptions and routing them 
based on the results. However, caching and content delivery are very important for publish/subscribe 
applications in which the objects of interest have large sizes. In this paper, we propose several content 
distribution approaches that combine match-based pushing and access-based caching, based on users'
subscription  information and access patterns. To study the performance of the proposed approaches,
we built a simulator and developed a workload to mimic the content and access dynamics of a busy
news site. We use two request traces: the trace NEWS uses a large homogeneity parameter for a 
Zipf-like page popularity distribution based on the observations at the news site, and the ALTERNATIVE
trace represents more general scenarios. The experimental results using our best approaches
demonstrate about 50% relative improvement in the hit ratio in local caches for NEWS and about 140%
for ALTERNATIVE over the baseline algorithm that uses access-based caching only. Even when the
subscription information is assumed not to reflect users' accesses perfectly (a more realistic scenario), 
our best approaches still have about 40% and 90% relative improvement for the two traces. In all cases,
our best approaches introduce only traffic overhead that is comparable to the access-based caching-only
approach. To the best of our knowledge, this work is the first effort to investigate content distribution under
the publish/subscribe paradigm. Based on it, we believe that not only can publish/subscribe systems
incorporate content distribution effectively, but that general caching and content distribution systems can
be substantially enhanced by using publish/subscribe information judiciously when it is available.

Latest update: 13 June 2003 - Questions and Comments about the Site: fmc@inf.ufg.br